Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!rutgers!bagate!cbmvax!daveh From: daveh@cbmvax.commodore.com (Dave Haynie) Newsgroups: comp.sys.amiga Subject: Re: A3000UX competition Message-ID: <16429@cbmvax.commodore.com> Date: 11 Dec 90 00:11:08 GMT References: <2300@lpami.wimsey.bc.ca> <4134.275a6088@cc.helsinki.fi> <14710@sdcc6.ucsd.edu> Reply-To: daveh@cbmvax.commodore.com (Dave Haynie) Organization: Commodore, West Chester, PA Lines: 50 In article <14710@sdcc6.ucsd.edu> cleland@sdbio2.ucsd.edu (Thomas Cleland) writes: >>WROONG. SLC is much slower than a 68040-NeXT. NeXTstation is supposed to >>be even faster than a SparcStation 1+. >Cosmic! I had wondered whether reports that the 68040 would >beat out many SPARC chips was just overzealous Amiga crowing. To read the "spec sheets" (actually, manufacturer's hype sheets; the spec sheets are only rated in nanoseconds), the SLC cranks out 12 MIPS, the 25MHz 68040 itself is capable of up to 20 MIPS (which says absolutely nothing about how fast a real 68030 computer will go, since the system design determines how much of the potential can actually be reached), and a 25MHz 68030 kicks up about 6 MIPS. However, those are chip numbers and may have little to do with real world performance. A good model for this is the Personal Workstation review this month of the IBM PowerStation systems. These little workstations have this supposedly amazing "America" chipset, IBM's second generation RISC effort, and at least at the time of their release, set the record for SPECmarks on a desktop system. The PW folks seem pretty good at cranking out benchmark numbers; they run CPU, FPU, and Disk tests, and good ones like SPEC and AIM. They found, for a single person running a single program, the RS6000s do kick some butt, on both hard disk (well, it was UNIX, and they compare it to crawlers like PCs and Macs as well as disk-wise real computers), integer, and floating point. When they run a multitasking benchmark like AIM, the thing looks more like a really expensive '386 system. That RISC design, which does get speed from being superscalar, is extremely sensitive to the software its running. More traditional RISC architectures, like the SPARC and the 68040 (which will act very much like a RISC in real benchmarks, I suspect), aren't quite so sensitive, but still only perform like the marketing folks claim when they have full pipelines and all. The 68030 doesn't hit the same peaks, but it's also very hard to slow down with multitasking. All of which means, don't pay too much attention to marketing hype like "MIPS" numbers. When reality sets in, I'm sure the 68040 will be a nice improvement over the 68030, but I don't think you're often going to see 3x-6x speed increases outside of bad benchmarks and perhaps some finely turned special purpose single threaded number crunching jobs. It also helps to know ahead of time why you're buying this computer and what you expect to do with it. Lots of us techies want faster machines, and a good percentage of that crowd may even have a real use for them. >Thom Cleland "It is easier -- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy "I can't drive 55" -Sammy Hagar